Classification of Wayang Kulit Using Canny Feature Extraction and Convolutional Neural Network Algorithm

Abstract

Wayang kulit is a part of Indonesian culture known to the Javanese people, but the younger generation often has difficulty recognizing the wayang characters they are looking for online because of inaccurate search results. One popular story is the Mahabharata, with the characters of the Five Pandavas: Puntadewa (Yudistira), Bima, Arjuna, Nakula, and Sadeva. Because puppet characters have similar shapes, curves, clothing, and colors, it is often difficult to distinguish and remember. This shows the need for technology to help recognize puppet characters more easily. This research aims to solve this problem by utilizing Deep Learning techniques in Computer Vision to classify puppet images. Canny's feature extraction technique and DenseNet-121 architecture are used to detect patterns in the puppet image and classify them into appropriate categories. The dataset used consisted of 1028 images divided into four categories: Arjuna, Bima, Nakula & Sadewa, and Puntadewa. The framework implemented is CRISP-DM, with the implementation using the Python 3.11 programming language, TensorFlow 2.14, and the Google Colab tool. The results of the model evaluation through the confusion matrix showed 93% accuracy, 93% precision, 93% recall, and 92% f1 score. With these results, it is hoped that technology can facilitate and increase accuracy in recognizing the character of puppet puppets.